IBM executive outlines agentic AI role in retail banking overhaul

Srini Bala, a partner and industry leader at IBM, argued that large retail banks can use agentic AI as middleware to improve Gen Z customer service without committing immediately to a multi-year, multimillion-dollar core system replacement. His case frames the technology as a way to cut operational friction, reduce dependence on call centres and deliver API-led services while banks continue longer-running modernisation programmes.

In a Finextra expert opinion post, Bala said many tier-one institutions still rely on complex systems of record built up through decades of mergers, acquisitions and tactical technology deployments. He pointed to monolithic COBOL-based core ledgers, separated customer information files and end-of-day mainframe batch processing as features that can slow real-time retail banking.

Those systems remain valuable for resilience and regulatory requirements, according to Bala, but they create constraints when banks try to add digital channels on top. He said basic chatbots and deterministic automation struggle when a customer request requires action across cards, deposits and customer data platforms, contributing to broken journeys and weaker straight-through processing.

Gen Z service expectations

Bala linked the technology challenge to the rise of Gen Z as a retail banking customer segment. He said younger customers compare banks with digital-native platforms and expect real-time balance visibility, embedded finance tools and personalised financial guidance.

For large institutions, replacing a core banking system in full can be high risk and capital intensive, Bala said. He argued that agentic AI offers a “wrap-and-renew” approach, placing an orchestration layer between customer-facing applications and legacy infrastructure rather than requiring an immediate overhaul of the ledger.

Agentic AI, as described by Bala, differs from a large language model that mainly generates text or robotic process automation that follows fixed scripts. He said an agentic system can identify customer intent, retain context, route requests by meaning and call multiple internal tools through fragmented APIs.

How the model would work

Bala used a fee-waiver scenario to describe the mechanics. If a customer asks why a non-sufficient funds fee was applied while a payroll credit is pending, the agent would separate the request into several parts: the fee event, the incoming automated clearing house credit and the waiver request.

The agent would then query the deposit system to check the fee code, review the clearing schedule for the pending credit and retrieve customer data from the bank’s customer relationship management system, according to Bala. If internal rules are satisfied, including customer value thresholds and liquidity coverage, the agent could start a straight-through workflow to memo-post a refund.

In Bala’s example, the result would be sent as a JSON payload through an API gateway to the mobile application, producing a contextual notification without requiring a human service agent.

Potential bank benefits

Bala said the same model could combine customer information from retail banking, cards and wealth management, giving the bank a more complete view across channels. He also said transaction data could be used to issue alerts about liquidity pressure and present contextual credit options such as buy-now-pay-later products or short-term overdraft lines.

The argument is an external opinion published by Finextra, which states that such posts are provided by their authors without editing and reflect the author’s views. Bala’s central claim is that agentic AI can give banks more time to carry out cards platform and cloud modernisation while improving the customer experience layer sooner.

This story draws on original reporting from Finextra Research.